Prathamesh Devadiga

Hanover, NH

I am an incoming CS PhD student at Dartmouth College, working with Prof. Shawn Shan on machine learning security and LLM privacy. I completed my B.Tech in Computer Science at PES University, Bangalore.

My research sits at the intersection of ML security, production AI systems, and trustworthy AI. The core question I keep coming back to: what does a model actually know, and what are the limits of our ability to control that?

I also lead Ādhāra AI Labs, an independent research lab working on compiler optimization with small language models, efficient neural architectures, and production LLM infrastructure. I was previously a research intern at Lossfunk and completed Google Summer of Code 2025 with UCSC-OSPO.

If you would like to chat about ML security research, collaborations, or graduate school, feel free to email me.

On this site, you can find my publications, projects, talks, and blog.

news

Sep 01, 2026	Starting my CS PhD at Dartmouth College in September 2026, with Prof. Shawn Shan, working on ML security and LLM privacy.
Jul 09, 2026	Paper accepted at ECCV 2026 UandME Workshop: Deletion Scars: Inferring Deleted Artist Styles from Paired Diffusion Models.
Jun 01, 2026	Paper accepted at ICML 2026 AI4Law Workshop: Resistant to Lawyers, Defeated by Disagreement.
Aug 01, 2025	Completed Google Summer of Code 2025 with UCSC-OSPO on billion-scale ANN embedding benchmarks.
Mar 01, 2025	Tulu structured prompting paper accepted at EACL 2025 LoResLM Workshop.
Jan 01, 2025	Research affiliate at Lossfunk, working on low-resource NLP and structured prompting.

latest posts

Sep 03, 2025	Advanced Jailbreak Attacks on Large Language Models: A Technical Analysis
Aug 05, 2025	ML System Load Testing and Capacity Planning
Jul 26, 2025	Real-time ML Inference: Serving Models with Low Latency

selected publications

ECCV

Deletion Scars: Inferring Deleted Artist Styles from Paired Diffusion Models

Prathamesh Devadiga

In ECCV 2026 Workshop on Unlearning and Model Editing (UandME), 2026

Bib HTML

@inproceedings{devadiga2026deletionscars,
  title = {Deletion Scars: Inferring Deleted Artist Styles from Paired Diffusion Models},
  author = {Devadiga, Prathamesh},
  booktitle = {ECCV 2026 Workshop on Unlearning and Model Editing (UandME)},
  year = {2026},
}

ICML

Resistant to Lawyers, Defeated by Disagreement: Evaluation Blindspots in Legal Language Models

Prathamesh Devadiga and Advika Lakshman

In ICML 2026 AI4Law Workshop, 2026

Bib

@inproceedings{devadiga2026lawyers,
  title = {Resistant to Lawyers, Defeated by Disagreement: Evaluation Blindspots in Legal Language Models},
  author = {Devadiga, Prathamesh and Lakshman, Advika},
  booktitle = {ICML 2026 AI4Law Workshop},
  year = {2026},
}

NeurIPS

SLMs as Compiler Experts: Auto-Parallelization for Heterogeneous Systems

Prathamesh Devadiga and others

In NeurIPS 2025 Workshop on Machine Learning for Systems, 2025

Bib

@inproceedings{devadiga2025slmcompiler,
  title = {{SLMs} as Compiler Experts: Auto-Parallelization for Heterogeneous Systems},
  author = {Devadiga, Prathamesh and others},
  booktitle = {NeurIPS 2025 Workshop on Machine Learning for Systems},
  year = {2025},
}

EACL

Can LLMs Learn Tulu? Teaching Low-Resource Languages Through Hard Constraint Prompting

Prathamesh Devadiga and others

In EACL 2026 LoResLM Workshop, 2026

Bib

@inproceedings{devadiga2025tulu,
  title = {Can {LLMs} Learn Tulu? Teaching Low-Resource Languages Through Hard Constraint Prompting},
  author = {Devadiga, Prathamesh and others},
  booktitle = {EACL 2026 LoResLM Workshop},
  year = {2026},
}